Uncertainty Quantification

Overview

We are working to develop a rigorous mathematical theory to quantify the variation in predictions made by stochastic models of epidemics. For these types of models, random events cause important factors, such as epidemic size and duration, to vary over a wide range of values. The goal is to find what quantities agent-based models predict well and what quantities will have large variation in their predictions.

The variation in model output makes it difficult for policy makers to base decisions on these types of epidemic models. Identifying the risk associated with a given prediction makes these large-scale simulations a more useful tool for the public health community.

In a small epidemic percolation network SIR simulation of 5,000 individuals the arrival time to the epidemic peak and the peak simultaneous percent infected are computed. These are computed for many different average transmission rates and many different contact networks. This yields a distribution of predictions graphed above. We notice that simulations resulting in earlier epidemic peaks have, on average, a higher percentage of individuals infected at the peak.